Search Process Overview

The search engine follows a structured eight-step process when handling a query:

Query Handling Process

Preprocessing the Search Term Before processing the search, the system prepares the term by:
- Removing special characters
- Normalizing terms (e.g., "Levi's" → "levis", "s.oliver" → "soliver")
- Removing hyphens and thousand separators
- Converting negation words (e.g., "without" → "-" prefix)
- Removing superscript numbers/letters
- Removing stop words (e.g., "the", "and", "also")
- Replacing umlauts
- Converting to lowercase
- Splitting compound words
- Applying stemming and spelling correction
If a user searches for "Schino pants", it is corrected to "Chino pants" and used as a synonym.
Checking for Direct Links If a predefined direct link exists for the query, the user is redirected to a:
- Content page
- Product detail page
- Specific internal or external URL If no link is found, the process continues.
Searching for Replacements The system checks for predefined replacements for the term. If found, they affect Step 7 (result merging). Regardless, search continues as normal.
Finding Synonyms The system checks for stored synonyms and adds them to the query. They are included with a small ranking penalty.

Example: A search for "Dirndl" includes "Country house dress", "Traditional costume", "Traditional dress".
Finding Synonyms for Misspellings Spellchecker-corrected terms are also added as synonyms. A penalty is applied based on the distance between the original and the corrected word.

Example: "Schino pants" → "Chino pants" (with a ranking penalty).

Searching the Search Index The search uses previously generated wordList and cutWordList. Every result starts at 100 points. Points are deducted based on where the term appears.

Data Field	Penalty	Description
Brand	0	Product brand
Short Description	0	Shown in result overview
Target Group	0	Men, Women, Children
Categories	4	Category placement
Product Group	4	Internal classification
Market Identifier (MKZ)	4	Internal marker
Primary Color	8	Main product color
Filter Values	8	Matched filterable attributes
Dimensions	8	Size categories like “Short”, “Normal”, “Long”
Long Description	40	Full text description

Merging Results (if applicable) If replacements were found in Step 3, multiple result sets are merged and re-ranked.

See Replacements for details TODO

Example: A replacement for "Ski Jacket" returns not only jackets, but also Ski Pants, Gloves, and Boots.
Outputting the Results Before the final results are returned:
- The product ranking is finalized
- Filters are calculated
- Results are serialized as JSON The structure can be tenant-specific
- Short-term caching applies only for large result sets

Ranking of Variations

Many products have multiple variants (e.g., color, size). The system ranks these variations using:

Rules from the Rule Engine
A BI-based scoring system to prioritize the best variant

Only the top-ranked variant is shown in the results.

Additional Factors in Search Ranking

Availability (Stock Level)

Stock Level	Penalty
≥ 80% available	No penalty
45% – 80% available	-10 points
< 45% available	-50 points
0% (out of stock)	Excluded entirely

Business Rule Value

Adds 5% weight to the product ranking
Values are normalized logarithmically
Top product gets a 100% boost, others scale accordingly